Is Violence Risk Assessment the same across gender? A Meta-analysis of the HCR-20 V3 in female forensic populations

Sapphire-Violet Rossdale, University of Nottingham
Ruth J. Tully, Tully Forensic Psychology Ltd.
Vincent Egan, University of Nottingham

This meta-analysis aimed to examine the effectiveness of the Historical Clinical Risk-20 violence management tool for predicting violence in adult female populations. The HCR-20 is a well-known tool which has been predominantly developed and validated on males. To date, there have been no published systematic reviews explicitly explore all generations of the HCR-20 and the predictive validity of the tool for females which encompasses measures of violence, recidivism and settings. Nine electronic databases and reference lists of relevant analyzes were searched. All papers were reviewed with predefined inclusion criteria, and quality assessment. Relevant papers were then subject to data extraction and synthesis. The electronic search revealed 1405 papers. After the exclusion process and quality assessment, 12 studies were included in the review. A meta-analysis was conducted using a total of 45 effect sizes assessing the predictive validity of the H, C, R, and Total items for violence and recidivism, followed by a sensitivity analysis. Results suggested a small effect size for H, C, and R items when assessing violence and for C, R, and total items for predicating recidivism in females. However, moderate effect sizes were found for total items when predicting violence and for H items for predicting recidivism. This review concludes that the HCR-20 assessment can be considered as a useful tool for understanding and assessing violence risk with females; however, further research is required to develop a more informed view of the HCR-20 assessment with females and should focus on the HCR-20 version three as the updated tool.

Risk Assessment, violence, females, historical clinical risk, HCR-20

”…the gender gap for occurrences of violence has been proposed to be narrowed in institutional settings, (e.g. forensic and civil psychiatric hospitals). Therefore, it is becoming increasingly important to understand risk and causes of violence committed by females. Risk assessment tools measuring likelihood of violence are important for prediction of future risk, and development of risk management plans.” (pp.15-16)

“It is suggested that females are more likely to engage in internalizing maladaptive behaviors, such as self-harm. Archer and Coyne found that females can use indirect aggression which would be toward themselves or those closest to them whilst masking the intent. Burbank explored 317 countries and suggested that female aggression is prevalent, but mostly expressed indirectly compared to males. Therefore, another explanation may be the nature of record keeping in forensic settings which may be amplifying the expressions of aggression in females as it is not typically recorded or considered in the community due to the “cultural ideal” and view that women are not typically violent, or do not express violence in the social constructed view of aggression.” (p.16)

“Overall, four reviews were found exploring violent risk assessment with females each of these are summarized below.
Geraghty and Woodhams (2015) examined the predictive validity of risk assessment tools for violent female offenders for 12 standardized violence risk assessment tools, with four studies assessing the HCR-20. The outcome of this suggested that the HCR-20 could predict violent recidivism for female offenders more accurately than other assessments. Geraghty and Woodhams (2015) concluded that the variability of AUC scores may be attributed to current risk assessment tools, not acknowledging specific risk factors associated with female offenders.

Singh et al. (2011) reviewed violence risk assessment tools and examined characteristics which may impact on the predictive validity of the measures, including gender. For the HCR-20, eight samples were used (n = 1320 participants). This review found the HCR-20 to have a good median AUC statistic (0.70; Interquartile Range [IQR] = 0.64 to0.76); however, a significant effect for gender was not found; suggesting that in this sample the HCR-20 is no more predictive for males than females.

Yang, Wong, and Coid (2010) conducted a meta-analysis on nine risk assessment tools, including the HCR-20, published between 1999–2008. This review found a moderate effect size among most of the tools. They concluded that heterogeneity was due to methodological differences (e.g. types of violent outcome, gender, length of follow up). This review aided in understanding that violence risk assessment tools are interchangeable, and assessment needs to be considered on an individual basis; however, it now lacks temporal validity.

Lastly, O’Shea, Mitchell, Picchioni and Dickens (2013) completed a review on the moderators of predictive efficacy for the HCR-20 for psychiatric inpatient violence. A meta-analysis was conducted on 20 studies and it was found that certain items on the HCR-20 (C5 – unresponsive to treatment and R5 – Stress) had larger effect sizes than others (H10 – Prior supervision failure) for predicting inpatient aggression, with the total risk summary outcome having the largest mean effect size. This review concluded that the HCR-20 had good predictive efficacy for females, showing some of the better predictive efficacy outcomes.” (

“Historical, clinical and risk management items for violence
For Violence, weak correlations were found for measuring violence for Historical, Clinical and Risk management items. A significant moderate correlation was found for the total items for predicting violence (k = 6, N = 493, r = .312, p < .001). For the Historical and Clinical items, the study most weighted revealed a considerably low effect size, which may have impacted on results. This particular study also revealed a high risk of bias and was therefore removed for a later sensitivity analysis.” (p.33)

“Historical, clinical and risk management items for recidivism
For the Historical, Clinical and Risk management and total items for assessing recidivism, a minimal number of studies was included in all analysis (k = 5). A significant medium effect size was found for the Historical items (k = 5, N = 537, r = .389, p < .001) when predicting recidivism. For the Clinical, Risk and Total items significantly low effect sizes were found.” (p.33)

“The implication for clinical practice is that the use of the HCR-20 may be appropriate for females but needs to be interpreted and applied with caution, and be considered with other sources, much like as has been advised by the authors. Assessors utilizing the tool in clinical practice need to remain aware of the lack of research into female risk and measurement of violence, particularly for specific subgroups including those with clinical diagnosis, intellectual ability, and levels of security.” (p.39)

“…female forensic services implement the HCR-20v3 as a standard assessment upon admission in the UK, despite the limited understanding of its ability to accurately assess female risk of violence. Recent developments in female risk assessment have suggested that the Female Additional Manual (FAM, de Vogel et al., 2014) is promising as a supplement to the HCR20 for assessing female violence accurately.” (p.39)

“Heterogeneity was significant when assessing all of the 12 included studies. Conducting meta-analysis in behavioral sciences can result in heterogeneity being inevitable due to subjective outcomes, such as violence. Variation in measurement of outcomes (violence or recidivism), differences in sample size (24 to 350), distinctive samples (civil psychiatric, forensic mental health and prisoners), versions of the HCR-20, and a combination of retrospective and prospective designs may have contributed to the lack of homogeneity … With regard to publication bias, this analysis suggests that publication bias was not a concern. However, it does not remove the risk of publication of positive results. Both journals and researchers may be concerned to suggest that a comprehensive tool, which has been implemented across services for 23 years, is actually not predictive of violence for a certain subgroup. The current small number of published studies which have assessed predictive validity of the HCR-20 in females suggests a possible lack of interest or ability to explore the area, or potentially risk of publication bias.” (p. 37-38)

As always, please join the discussion below if you have thoughts or comments to add!